Professional Documents
Culture Documents
Rieszs Approach to
the Lebesgue Integral
Douglas S. Bridges
Department of Mathematics & Statistics,
University of Canterbury,
Private Bag 4800,
Christchurch, New Zealand.
January 12, 2006
1
Preface
In the following, I introduce the Lebesgue integral on the real line R using
the method of F. Riesz. Working with increasing sequences of step functions
whose integrals are uniformly bounded above, this method, which is essentially a
special case of the Daniell approach to abstract integration, avoids the somewhat
tedious technical detail about measures that is required in the standard measure-
theoretic introductions to the Lebesgue integral, and thereby enables us rapidly
to reach the key results about convergence of sequences and series of integrable
functions.
The later sections of the notes contain material about the spaces Lp (R) of
ppower integrable functions on R; a development of the Lebesgue double inte-
gral, including Fubinis theorem about the equivalence of double and repeated
integrals; and a discussion of topics in advanced dierentiation theory, such
as Fubinis series theorem and the Lebesgue-integral form of the fundamental
theorem of calculus.
2
1 Compactness
Many results in real-variable theory depend on two fundamental properties of
the real line described in this introductory section.
By[a cover of a subset S of R we mean a family U of subsets of R such that
S U; we then say that S is covered by U and that U covers S: If also
each U 2 U is an open subset of R, we refer to U as an open cover of S (in
R). By a subcover of a cover U of S we mean a family F U that covers S;
if also F is a nite family, then it is called a nite subcover of U.
Although there exist shorter proofs of the next theorem (see the next set
of exercises), the one we present is readily adapted to work in a more general
context.
3
Proof. See Exercise (1.1, 2) below.
Exercises (1.1)
A = fx 2 I : S \ ( 1; x) is nite or emptyg
is a limit point of S in I:
.4 Let S be a subset of R with the BolzanoWeierstrasz property, and let
(xn )n>1 be a sequence of points of S: Prove that there exists a subsequence
of (xn ) that converges to a limit in S:
.5 Let S be a subset of R with the BolzanoWeierstrasz property. Prove that
S is closed and bounded. (For boundedness, use a proof by contradiction.)
.6 Show that the BolzanoWeierstrasz theorem can be proved as a conse-
quence of the HeineBorelLebesgue theorem. (Let I be a compact inter-
val in R; and suppose that there exists an innite subset S of I that has
no limit point in I: First show that for each s 2 S there exists rs > 0 such
that S \ (s rs ; s + rs ) = fsg.)
.7 Let f be a real-valued function dened on an interval I. We say that f
is uniformly continuous on I if to each " > 0 there corresponds > 0
such that jf (x) f (x0 )j < " whenever x; x0 2 I and jx x0 j < : Show
that a uniformly continuous function is continuous. Give an example of I
and f such that f is continuous, but not uniformly continuous, on I:
4
.8 Use the HeineBorelLebesgue theorem to prove the uniform continuity
theorem: a continuous real-valued function f on a compact interval I
R is uniformly continuous. (For each " > 0 and each x 2 I; choose x > 0
such that if x0 2 I and jx x0 j < 2 x ; then jf (x) f (x0 )j < "=2: The
intervals (x x ; x + x ) form an open cover of I:)
.9 Prove the uniform continuity theorem (see the previous exercise) using the
BolzanoWeierstrasz theorem. (If f : I ! R is not uniformly continuous,
then there exists > 0 with the following property: for each n 2 N+ there
exist xn ; yn 2 I such that jxn yn j < 1=n and jf (xn ) f (yn )j > :)
Let
c = 1 + max fjf (x1 )j ; : : : ; jf (xN )jg ;
and consider any point x 2 I: Choosing k such that x 2 (xk xk ; xk + xk ); we
have
so f is bounded on I:
Now write
m = inf f; M = sup f:
Suppose that f (x) 6= M; and therefore f (x) < M; for all x 2 I: Then x
1=(M f (x)) is a continuous mapping of I into R+ ; and so, by the rst part of
this proof, has a supremum G > 0: For each x 2 I we then have M f (x) >
1=G and therefore f (x) 6 M 1=G: This contradicts our choice of M as the
supremum of f:
5
Exercises (1.2)
Note that the value of f at xk is not constrained in any way. The integral of
this step function is dened to be
Z n
X1
f= ck jIk j ;
k=0
where jIk j = xk+1 xk denotes the length of the interval (xk ; xk+1 ) :
The proof is left as an exercise. It follows from this proposition that the
integral of a step function is well dened, in that its value does not depend on
the choice of the nitely many points xk that are the endpoints of intervals on
which the function is constant.
6
Moreover, f g; max ff; gg ; min ff; gg ;
f + = max ff; 0g ;
and
f = max f f; 0g
are step functions.
R Finally,
R if f (x) 6 g(x) for all but nitely many (possibly no)
points x, then f 6 g:
Exercises (2.1)
.1 Prove Proposition 4.
.2 Prove Proposition 5.
P (x) a:e:
If S is a subset of R; A is a null set; and P (x) is a property that holds for all x
in SnA; then we say that P (x) holds almost everywhere in S; and we write
P (x) a.e. in S:
Lemma 6 Let A be a null set. Then for each " > 0 there exists a Riesz sequence
(fn ) of nonnegative
R step functions such that fn (x) " 1 for each x 2 A; and such
that 0 6 fn 6 " for each n:
Proof. Choose a Riesz sequence (gn ) of step functions and M > 0 such that
7
(i) gn (x) " 1 for all x 2 A and
R
(ii) gn 6 M for all n:
Then (gn g1 )n>1 is a Riesz sequence of nonnegative step functions such that
R R
gn (x) g1 (x) " 1 for each x 2 A; and such that (gn g1 ) 6 M + jg1 j for
each n: It remains to take
"
fn = R (gn g1 ) :
M + jg1 j
Proof. Let (An ) be a sequence of null sets. By the preceding lemma, for
(n)
each n there exists a Riesz sequence fk of step functions that diverges
k>1
R (n)
to 1 on An and that satises fk 6 2 n for all n and k: Dene step functions
(1) (2) (k)
fk = fk + fk + + fk (k > 1) :
Then
(1) (k)
fk (x) = fk (x) + + fk (x)
(1) (k) (k+1)
6 fk+1 (x) + + fk+1 (x) + fk+1 (x) = fk+1 (x)
and Z
fk 6 2 1
+2 2
+ +2 k
< 1:
Proof. For each n there is a nite (possibly empty) set Dn of points where
fn is not continuous. Then [
D= Dn
n>1
is countable, say
Dn = fx1 ; x2 ; x3 ; : : :g :
8
We may assume that the endpoints of I belong to D: Choose M > 0 such that
jf1 (x)j 6 M for all x 2 R: Given " > 0; for each n let Jn be an open interval of
1 [ P1
length 6 2n+1 M " that contains xn : Then D Jn and jJn j 6 "=M:
n>1 n=1
Next, given x 2 InD; choose a positive integer x such that fn (x) 6 "=2 jIj
for all n > x : Since x 2 = D; f x is continuous at x and hence, being a step
function, is constant in some open subinterval Vx of (the interior of) I containing
x: Thus for all t 2 Vx and all n > x ;
"
fn (t) 6 f x (t) 6 :
2 jIj
The open intervals Jn (n > 1) and Vx (x 2 InD) form an open cover of I; from
which, by Theorem 1, we can extract a nite subcover, consisting, say, of the
intervals Jn (1 6 n 6 N ) and V k (1 6 k 6 m) : Dene step functions g1 ; g2 as
follows:
8
>
> S
N
< M if x 2 Jn
g1 (x) = n=1
>
>
:
0 otherwise,
8
> S
m
>
< "=2 jIj if x 2 Vk
k=1
g2 (x) =
>
>
:
0 otherwise.
Then
Z N
X 1
X 1
X "
g1 6 M jJn j 6 M jJn j 6 M
1
(2n M ) "=
n=1 n=1 n=1
2
and, as each V k
is in the interior of I;
Z m
" [ "
g2 6 V k
6 :
2 jIj 2
k=1
Let
= max f 1
;:::; m
g:
Then for each n > , since fn (x) 6 g1 (x) + g2 (x) for each x 2 I; we have
Z Z Z
fn 6 g1 + g2 6 ":
9
Proof. Choose a null set A such that limn!1 fn (x) = 0 for all x 2 RnA:
Given " > 0; use LemmaR 6 to nd a Riesz sequence (gn ) of step functions such
that gn > 0 and 0 6 gn 6 " for each n; and such that gn (x) " 1 for each
x 2 A: Consider the step functions
+
hn = (fn gn ) :
For each n;
for all su ciently large n; whence hn (x) = 0 for all su ciently large n: It now
follows
R that hn (x) ! 0 for all x 2 R: We can now apply Proposition 8 to show
that hn ! 0: Since
Z Z Z Z
fn = (fn gn ) + gn 6 hn + ";
R
it followsR that limn!1 fn 6 ": Since " > 0 is arbitrary, we conclude that
limn!1 fn = 0:
Let E " denote the set of all functions that are limits almost everywhere of
some Riesz sequence (fn ) of step functions. If f 2 E " ; and (fn ) is a Riesz
sequence of step functions such that f (x) = limn!1 fn (x) a:e:; then we dene
the (Lebesgue) integral of f to be
Z Z
f = lim fn :
n!1
Exercises (2.2)
.1 Prove that (i) any nonempty nite set and (ii) the set of rational numbers
is a null set. Is the empty subset of R a null set (explain your answer)?
.2 Let (fn ) and (gn ) be Riesz sequences of step functions such that
10
almost everywhere. Prove that
Z Z
lim fn > lim gn :
n!1 n!1
+
R(Consider the step functions m;n = (gm fn ) : Show that for each m;
m;n ! 0 as n ! 1:)
and
(ii) if f is a step function, then its Lebesgue integral coincides with the
integral as we originally dened it.
Now let
L = f1 f2 : f1 ; f2 2 E " :
The elements of L are called (Lebesgue) summable functions. If f1 ; f2 2 E "
and f = f1 f2 ; dene the (Lebesgue) integral of f to be
Z Z Z
f = f1 f2 :
Exercises (2.3)
11
.3 Let f; g be summable functions.
R Prove
R that max ff; gg, min ff; gg ; and
jf j are summable, and that f 6 jf j : (First consider the case where
f; g are both nonnegative.)
.4 Prove that if f is summable, then the functions
1
min ff; ng ; max fmin ff; ng ; ng ; and min jf j ;
n
are summable for each positive integer n:
.5 Prove that the Lebesgue integral is translation invariant: if f 2 L and
yR 2 R; then
R the function fy dened by fy (x) = f (x + y) is summable, and
fy = f: (Reduce to the case where f 2 E " :)
.6 Prove that if f is summable and is a nonzero
R real
R number, then h(x) =
f ( x) denes a summable function and j j h = f:
Note that
12
Proposition 10 If f is summable over the subsets A; B of R; then it is sum-
mable over A [ B and A \ B:
both of which are summable by Exercise (2.3, 3). In the general case, set f =
f + f : Then f + 2 L; by Exercise (2.3, 3). Since f + A = max ff A ; 0g ; the
same exercise shows that f + is summable over A; similarly, f + is summable
over B: It follows from the rst part of this proof that f + is summable over
both A [ B and A \ B: Similarly, f is summable over A [ B and A \ B: Hence
Exercises (2.4)
Proposition 11 Let (fn ) be a sequence of elements of E " such that fn (x) > 0
1 R
P P
1
a.e. and fn converges in R: Then there exists f 2 E " such that fn (x) =
n=1 n=1
R 1 R
P
f (x) a.e. and f= fn :
n=1
(n)
Proof. For each n choose1 a Riesz sequence fk of step functions
k>1
and a null set An such that
(n)
(i) limk!1 fk (x) = fn (x) for all x 2 RnAn ;
R (n) R
(ii) limk!1 fk = fn ; and
(n)
(iii) fk > 0 for each k:
1 Thisrequires a little thought, though not much. Let (gn )n>1 be a Riesz sequence of step
functions converging almost everywhere to f; and take fn = max f0; gn g :
13
Dene the step functions
(1) (2) (i)
gi = fi + fi + + fi (i > 1) :
(n)
Then (cf. the proof of Proposition 7) gi 6 gi+1 : Also, since fi " fn as i ! 1;
Z Z Z Z X 1 Z
gi 6 f1 + f2 + + fi 6 fn :
n=1
Hence (gi )i>1 is a Riesz sequence of step functions, and so there exist f 2 E "
R R
and a null set A0 such that gi (x) " f (x) for all x 2 RnA0 and gi " f as
i ! 1: Let
1
[
A= An ;
n=0
which, being a countable union of null sets, is a null set. For each x 2 RnA;
f (x) > gi (x)
(1) (2) (i)
= fi (x) + fi (x) + + fi (x)
(1) (2) (n)
> fi (x) + fi (x) + + fi (x) (n 6 i) :
Keeping n xed and letting i ! 1; we obtain
f (x) > f1 (x) + + fn (x):
P
1
So the series fn (x) has partial sums bounded above by f (x):Since
n=1
(n)
fn (x) = lim fi (x) > 0 (x 2 RnA) ;
i!1
that series converges to a sum at most f (x) for each x 2 RnA: For each such x;
gn (x) 6 f1 (x) + + fn (x) 6 f (x) (n > 1) ;
so
1
X
f (x) = lim gn (x) 6 fn (x) 6 f (x)
n!1
n=1
and therefore
1
X
f (x) = fn (x):
n=1
Finally,
Z i Z
X Z
gi 6 fn 6 f;
n=1
R 1 R
P
so, letting i ! 1; we have f= fn :
n=1
14
Lemma 12 Let f 2 L; and let f (x) > 0 a.e. Then for each " R> 0 there exist
g; h 2 E " such that f = g h; g(x) > 0 a.e, h(x) > 0 a.e., and h < ":
Proof. Choose g (1) ; h(1) in E " such that f = g (1) h(1) : There exists a
Riesz sequence (hn ) of step functions such that
R
R
Given " > 0; choose N such that h(1) hN < ": Take h = h(1) hN and
(1)
g=g hN : Since hNR is a step function, so is hN ; and therefore both g and
h belong to E " : Also, h < "; h(x) > 0 a.e., and g(x) = f (x) + h(x) > 0 a.e.
Finally, f = g h a.e.
Proof. Using Lemma 12, choose sequences ( n ) ; ( n ) inR E " such that for
n ; n (x) > 0 a.e., n (x) > 0 a.e., and 0 6 n 62
n
each n; fn = n : Then
PR
1 P
1
the series n converges by comparison with 2 n : Hence, by Proposition
n=1 n=1
P
1 P1 R R
"
11, there exists2 E such that n (x) = (x) a.e. and n = :
R R n=1 R n=1
Since n = fn + n , n = fn + n and the convergence of the series
P1 R 1 R
P P1 R
fn ; n entails that of n : Applying Proposition 11 once more,
n=1 n=1 n=1
P
1 1 R
P R
"
we obtain 2 E such that n (x) = (x) a.e. and n = : Since
n=1 n=1
the union of two null sets is a null set, it remains to take f = :
Exercises (2.5)
15
.3 Prove that a null set cannot contain a proper interval.
.4 Let (An ) be a sequence of subsets of R, and f a function that is summable
P1 R
over each An ; such that An
jf j converges. Prove that
n=1
[ R 1 R
P
(i) f is summable over A = An ; and A
jf j 6 An
jf j ;
n>1 n=1
R 1 R
P
(ii) if also the sets An are pairwise disjoint, then A
f= An
f:
n=1
We are now in a position to show that the extension process that took us
from the set of step functions to E " and then L cannot be applied, beginning
with L; to extend the class of summable functions.
gn = fn+1 fn (n > 1) :
P
k
Then gn (x) > 0 a.e. and gn = fk+1 f1 ; so
n=1
k Z
X Z Z
gn = fk+1 f1 6 M (k > 1):
n=1
1 R
P
Thus the series gn of nonnegative terms has bounded partial sums and so
n=1
P
1
is convergent. Applying Theorem 13, we obtain g 2 L such that gn (x) =
n=1
P
1 R R
g(x) a:e: and gn = g: Now let f = g + f1 2 L: Then
n=1
1
X
lim fn (x) = f1 (x) + gn (x) = f1 (x) + g(x) = f (x) a:e:
n!1
n=1
16
and
Z Z 1 Z
X Z Z Z Z
lim fn = f1 + gn = f1 + g= (f1 + g) = f;
n!1
n=1
as we required.
Corollary 15 Let (fn ) be a sequence of summable functions such that f1 >
f2 > > 0 a.e.
R Then
R there exists a summable function f such that fn (x) !
f (x) a.e. and fn ! f as n ! 1:
Proof. Apply Beppo Levis theorem to the sequence ( fn )n>1 of summable
functions.
Exercises (2.6)
.1 Dene f (x) = e x ; where is a positive
R constant. Prove that f is
summable over [0; 1); and calculate f: (You may assume something
that we will prove in Section 4: namely, that if f : [a; b] ! R is Riemann
integrable, then it is summable and the Lebesgue integral of f equals its
Riemann integral.)
P
1
n2 x
.2 Prove that the series e converges for each x > 0: Dene
n=1
8 1
> P n2 x
>
< e if x > 0
n=1
f (x) =
>
>
: 0 if x 6 0:
R P
1
Prove that f is summable and that f= 1=n2 :
n=1
17
1
.6 Let (fn )n=0 be a sequence of nonnegative summable functions such that
fn (x) 6 f0 (x) a.e. for each n: Prove that supk>n fk is summable for each
n; that
lim sup fn = lim sup fn
k!1 n>k
gn = sup fk
k>n
hn = inf fk :
k>n
Finally, hn 6 fn 6 gn ; so
Z Z Z
hn 6 fn 6 gn
R R
and therefore fn ! f:
Exercises (2.7)
R R
.1 Prove that if f is summable, then min ff; ng ! f as n ! 1. (Note
that min ff; ng is summable, by Exercise (2.3, 4).)
18
.2 Let f be a summable function, and for each positive integer n dene
1
x n f x+ f (x) ;
n
where n is a positive integer. Once again, you will have to anticipate the
connection between Riemann and Lebesgue integrals. Also, you will need
to use some results from elementary calculus.)
19
Proof. We prove if; the converse is left as an exercise. Accordingly,
assume (i) and (ii), and choose a null set A1 such that fn (x) ! f (x) for all
x 2 RnA1 : Compute a strictly increasing sequence (nk )k>1 of positive integers
such that for each k;
Z
jfm fn j < 2 k (m; n > nk ) : (1)
R k
1 R
P
Then fnk+1 fnk < 2 and so the series fnk+1 fnk converges, by
k=1
P
1
k
comparison with 2 : It follows from Lebesgues series theorem that there
n=1
P
1
exist a summable function and a null set A2 such that fnk+1 (x) fnk (x)
k=1
converges to (x) for all x 2 RnA2 : Then A = A1 [ A2 is a null set, and for all
P
1
x 2 RnA the series fnk+1 (x) fnk (x) converges and
k=1
1
X
fn1 (x) + fnk+1 (x) fnk (x) = lim fnk (x) = f (x):
k!1
k=1
20
Exercises (2.8)
R
.1 Under the hypotheses of Proposition 17, prove that jf fn j ! 0 as
n ! 1:
.2 Prove the following converse of Proposition 17: If f is summable,
R then
there exists a sequence (fn ) of step functions such that jf fn j ! 0
and fn (x) ! f (x) a.e.
1
An = x 2 R : fn (x) > (n > 1) :
"
Clearly, [
A An and An An+1 (1)
n>1
Also, since each fn is a step function, each An is the union of a nite number
of disjoint (possibly degenerate) intervals. For each of the intervals I that make
up An in this way we have I 6 "fn ; so the total length of these intervals is
Z
jAn j 6 " fn 6 ":
Next, dene
B1 = A1 ; Bn = An nAn 1 (n > 2) :
Then Bn is a nite union of disjoint (possibly degenerate) intervals, and
k
[
Bn = Ak (k > 1) :
n=1
Hence, by (1), [ [
Bn = An A:
n>1 n>1
21
Further, the total length of the disjoint subintervals that compose Bn is
so
k
X
jBn j = jAk j 6 " (k > 1)
n=1
P
1
and therefore jBn j converges to a sum 6 ": Since " > 0 is arbitrary, we
n=1
conclude that A has measure zero.
k = min f A1 ; : : : ; Ak g ;
S
i
Now let n;i be the characteristic function of Jn;k , which is a step function.
k=1
We have
n;i 6 n;i+1 (i > 1) ;
lim n;i (x) = An (x) (x 2 R) ;
i!1
and
Z i
X
n;i 6 jJn;k j 6 1=n (i > 1) :
k=1
22
Hence ( n;i )i>1 is a Riesz sequence of step functions converging almost every-
where to An ; so An 2 E " and
Z Z
1
06 An = lim n;i 6 :
i!1 n
It follows that n is summable and that
Z Z
1
06 n 6 An 6 :
n
Since also k > k+1 ; we can apply Corollary 15 to show, with reference to (1),
that S 2 L and that
Z Z
06 S = lim k = 0:
k!1
The following exercises are designed to show that not all null sets are count-
able.
Exercises (3.1)
.1 Let C be the Cantor set that is, the subset of [0; 1] consisting of all
P1
numbers that have a ternary (base 3) expansion an 3 n with an 2
n=1
f0; 2g for each n: Prove that if a; b are two elements of C that dier in
their mth ternary places, then ja bj > 3 m :
.2 With C as in the previous exercise, prove that [0; 1] nC is the union of a
sequence (Jn )n>1 of non-overlapping open intervals whose lengths sum to
1; and that C has measure zero.
P
1
n
.3 For each x = an 3 2 C dene
n=1
1
X
n 1
F (x) = an 2 :
n=1
23
4 Riemann and Lebesgue Integrals
We now recall the basic denitions of Riemann integration theory. By a parti-
tion of a compact interval I = [a; b] we mean a nite sequence P = (x0 ; x1 ; : : : ; xn )
of points of I such that
a = x0 6 x1 6 6 xn = b:
are called the lower sum and upper sum, respectively, for f and P: Since
exist. Moreover,
Z b Z b
f6 f: (1)
a a
the partition P:
24
so if Q is any partition, then
Thus every lower sum for f is less than or equal to every upper sum, from which
(1) follows.
We say that f is Riemann integrable over I if its lower and upper integrals
coincide, in which case we dene the Riemann integral of f over I to be
Z b Z b Z b
f= f= f:
a a a
R
We also dene Z Z
a b
f = f
b a
R R
and Z
1
(hPn gPn ) < :
n
25
Dene step functions
and Z Z
1
06 ( n n) 6 (hPn gPn ) < : (2)
n
It follows from (1) and (2) that
Z Z a Z
lim n = f = lim n:
n!1 b n!1
R
R R
Since n 6 hP1 for each n; we see that ( n ) is a Riesz sequence of step
functions; whence there exist a null set A and a function 2 E " such that
limn!1 n (x) = (x) for each x 2 RnA; and
Z Z
lim n = : (3)
n!1
06 n+1 n+1 6 n n;
R
and ( n n ) < 1=n: By Exercise (2.6, 7), there exists a null set C such that
limn!1 ( n (x) n (x)) = 0 for all x 2 RnC: Then
S =A[B[C
and therefore f (x) = (x): Thus f is summable (in fact, both f and f are in
E " and, by (3), Z Z Z Z a
f= = lim n = f:
n!1 b
R
26
The converse of Theorem 20 is false. Dene f : [0; 1] ! f0; 1g by
0 if x 2 [0; 1] \ Q
f (x) =
1 if x 2 [0; 1] nQ:
Rb Rb
Then a f = 0 and a f = 1; so f is not Riemann integrable. However, extending
f by setting f (x)
R = 0 if x < 0 or x > 1; we see that f (x) = 0 a.e.; whence f is
summable and f = 0:
Exercises (4.1)
.3 Let F be continuous on R; and let a < b: Prove that for each x 2 [a; b] ;
R x+ 1
n x n F ! F (x) as n ! 1:
.4 Let f and fn (n > 1) be Riemann integrable over [a; b] : Suppose that
fn 6 fn+1 for each n; and that limn!1 fn (x) = f (x) for each x 2 [a; b] :
Rb Rb
Prove that a fn ! a f as n ! 1: (Note that we need to include the
R R
hypothesis that f is Riemann integrable here; without it, we could prove
is that f is summable, but not necessarily that it is Riemann integrable.)
These are well dened. For example, if > 0; then for each x 2 I; as f is
bounded,
M (x) = sup ff (t) : t 2 I \ (x ; x + )g
exists; moreover, M (x) is an decreasing function of and is bounded below by
inf ff (t) : t 2 Ig : Hence f (x) = lim #0 M (x) exists; a similar argument shows
that f (x) exists.
27
Lemma 21 Let f : [a; b] ! R be bounded, and x0 2 [a; b] : Then f is continuous
at x0 if and only if f (x0 ) = f (x0 ) = f (x0 ):
Proof. Exercise.
Then
To see this, let " > 0 and choose " > 0 such that if 0 < < " ; then 0 6
M (x) f (x) < ": There exists a positive integer N such that mesh (Pn ) < "
28
for all n > N: Let n > N; and let (xk 1 ; xk ) be the partition interval for Pn
that contains x: Then (xk 1 ; xk ) (x " ; x + " ) ; so
and therefore
and
Z Z Z b
f = lim Pn = f:
n!1 a
Exercises (4.2)
29
Proposition 24 Integration by Parts: Let f; g be summable over the com-
pact interval [a; b] ; and let
Z x Z x
F (x) = f; G(x) = g (a 6 x 6 b) :
a a
Proof. In the trivial case where f (x) = c1 and g(x) = c2 are constant for
all x 2 [a; b] ; we use Theorem 20 to show that
Z b Z b Z b
2
Fg = c1 c2 (x a) dx = 21 c1 c2 (b a) = fG
a a a
R
The desired conclusion now follows in the case where f; g are step functions
that vanish outside [a; b] : In the case of general summable functions f; g we use
Proposition 17 and Exercise (2.8, 1) to construct sequences R (fn ) ; (gn ) of step
functions
R such that fn (x) ! f (x) a.e., gn (x) ! g(x) a.e., jf fn j ! 0; and
jg gn j ! 0 as n ! 1: For each n write
8 Rx
< a fn if a 6 x 6 b
Fn (x) = ;
:
0 otherwise,
8 Rx
< a gn if a 6 x 6 b
Gn (x) = :
:
0 otherwise.
By (an obvious variant of) Exercise (2.7, 5), Fn and Gn are both continuous,
and hence Riemann integrable, on [a; b] : Since gn is a step function and therefore
continuous a.e., the function Fn gn is Riemann integrable, and hence summable
(Theorem 20), over [a; b] : Similarly, fn Gn is summable over [a; b] : Now, for each
x 2 [a; b] ;
Z x
jF (x) Fn (x)j = (f fn )
Z ax Z
6 jf fn j 6 jf fn j ! 0 as n ! 1
a
and Z x Z
jFn (x)j = fn 6 jfn j ;
a
30
with similar inequalities holding for jG(x) Gn (x)j and jGn (x)j : Thus
jF (x)g(x) + f (x)G(x) (Fn (x)gn (x) + fn (x)Gn (x))j
6 jF (x) Fn (x)j jg(x)j + jFn (x)j jg(x) gn (x)j
+ jf (x)j jG(x) Gn (x)j + jfn (x) f (x)j jGn (x)j
Z Z Z
6 jg(x)j jf fn j + jg(x) gn (x)j jfn j + jf (x)j jg gn j
Z
+ jf (x) fn (x)j jgn j
and so
Z b Z b
(F g + f G) (Fn gn + fn Gn )
a a
Z Z Z Z
6 jgj jf fn j + jg gn j jfn j
Z Z Z Z
+ jf j jg
gn j + jf fn j jgn j
Z Z Z Z
= jgj + jgn j jf fn j + jf j + jfn j jg gn j :
Thus Z Z
b b
(F g + f G) (Fn gn + fn Gn ) ! 0 as n ! 1;
a a
and so
Z b Z b
(F g + f G) = lim (Fn gn + fn Gn )
a n!1 a
= lim Fn (b)Gn (b) (by the step-function case)
n!1
= F (b)G(b):
31
5 Measurable functions
Let f be a real-valued function dened almost everywhere on R: We say that f
is measurable if it is the limit a.e. of a sequence of step functions.
Exercises (5.1)
Then gn is summable, by Exercise (2.3, 3), and for each x 2 RnA; jgn (x)j 6 g(x)
and
lim gn (x) = max f g(x); min ff (x); g(x)gg = f (x):
n!1
Exercises (5.2)
32
.1 Prove that a measurable function f is summable if and only if jf j is sum-
mable.
.2 Prove that a summable function is summable over any interval.
.3 Let (fn ) be a sequence of measurable functions that converges almost
everywhere to a function f: Prove that f is measurable. (For each k
dene the step function gk by
8
< k if k 6 x 6 k
gk (x) =
:
0 otherwise.
First prove that max f gk ; min ff; gk gg is summable.)
p
.4 Let f be measurable, and p a positive number. Prove that jf j is measur-
able.
.5 Give an example of a measurable function f which is not summable even
though f 2 is summable.
.6 Give two proofs that a product of measurable functions is measurable.
(For one proof use Exercises (5.1, 2) and (5.2, 5).)
.7 Let f be measurable, and nonzero almost everywhere. Prove that 1=f is
measurable. (First consider the case where f > c almost everywhere for
some positive constant c: For general f > 0; consider fn = 1= f + n 1 :)
.8 Let 1 < < 2; and dene
8
< x sin x if x > 0
f (x) =
:
0 if x 6 0:
Prove that f is summable.
.9 Prove the RiemannLebesgue lemma: If f is a summable function,
then the functions x f (x) sin nx and x f (x) cos nx are summable,
and
Z Z
lim f (x) sin nx dx = 0 and lim f (x) cos nx dx = 0:
n!1 n!1
33
.11 Let f be a summable function, and g a bounded summable function such
that limx! 1 g(x) = 0 = limx!1 g(x) almost everywhere. Prove that
the function h : x R g(x)f (x=n) is summable for each positive integer n;
and that limn!1 n1 h = 0: (Hint: See Exercise (2.3, 5).)
Proof. Choose a sequence (fn ) of step functions, and a null set A; such that
fn (x) ! f (x) for all x 2 RnA: For each n dene
gn = fn (0):
Then gn is a step function and so is measurable. On the other hand, the constant
function x (0) is measurable (why?); so fn is measurable, as the dierence
of two measurable functions, by Exercise (5.1, 2). The continuity of ensures
that fn (x) ! f (x) for each x 2 RnA: It follows from Exercise (5.2, 3)
that f is measurable.
34
Exercises (5.3)
The elements of B are called Borel sets. Prove that any Borel set is
measurable.
35
Proof. Suppose that f is measurable, let r 2 R, and for each positive
integer n dene
(f r)+
fn = 1 :
n + (f r)+
Since the functions t t+ and
t
t 1
n +t
are continuous on R and R0+ ; respectively, we see from Exercises (5.1, 3) and
Proposition 27 that fn is measurable. But
:::;r 2 ; r 1 ; r0 ; r1 ; r2 ; : : :
n
such that 0 < rk+1 rk < 2 for each k: Then
is measurable, by Exercise (5.3, 1); let k denote its characteristic function. The
function
1
X
fn = rk 1 k
k= 1
The exercises in the next set extend the ideas used in the proof of Proposition
28.
Exercises (5.4)
(i) f is measurable.
36
(ii) [[f > r]] is measurable for each r.
(iii) [[f 6 r]] is measurable for each r:
(iv) [[f < r]] is measurable for each r:
(v) [[r 6 f < R]] is measurable whenever r < R:
.2 In the notation of the second part of the proof of Proposition 28, prove
that if fR is nonnegative
R and summable, then each fn is summable and
limn!1 fn = f:
.3 Let f be a nonnegative measurable function vanishing outside the interval
[a; b]. For the purpose of this exercise, we call a sequence (rn )1
n=0 of real
numbers admissible if r0 = 0 and there exists > 0 such that rn+1 rn <
P
1
for all n; and we say that the series rn (En ) corresponds to the
n=1
admissible sequence, where En ; whose characteristic function we denote
by n ; is the measurable set [[rn 1 6 f < rn ]]: Suppose that this series
converges. Let (rn0 )1 0
n=0 be any admissible sequence for f; and let n be
the characteristic function of En = [[rn 1 6 f < rn ]]. Prove that
0 0 0
P
1 P
1
(i) the series rn0 0
1 n and rn n converge almost everywhere to
n=1 n=1
summable functions,
P
1 P
1
(ii) rn0 1 0n 6 f 6 rn n almost everywhere,
n=1 n=1
P
1 P
1
(iii) the series rn0 1 (En0 ) and rn (En ) converge, and
n=1 n=1
P
1 P
1
(iv) rn0 1 (En0 ) 6 rn (En ):
n=1 n=1
P
1
Hence prove that if rn (En ) converges for at least one admissible
n=1 R
sequence (rn ); then f is summable, and f is both the inmum of the set
(1 )
X
rn (En ) : (rn ) is admissible, 8n (En = [[rn 1 6 f < rn ]])
n=1
37
P
1 P
1
sums analogous to rn (En ) and rn 1 (En ); and showed that the
n=1 n=1
inmum of sums of the rst type equals the supremum of sums of the
second; the common value of the inmum and supremum is precisely the
Lebesgue integral of f:)
.4 By a simple function we mean a nite sum of functions of the form c ;
where c 2 R and is the characteristic function of an integrable set. Let
f be a nonnegative summable function. Show that there exists a sequence
(fn ) of simple functions such that
Are all subsets of R measurable? No: the axiom of choice (Appendix) ensures
that non-measurable sets exist.3 To show this, following Zermelo, we dene an
equivalence relation on (0; 1) by
x y if and only if x y 2 Q:
Let x_ denote the equivalence class of x under this relation. By the axiom of
choice, there exists a function on the set of these equivalence classes such that
(x)
_ 2 x_ (x 2 (0; 1)):
3 Solovay has shown that there is a model of Zermelo-Fraenkel set theory, without the
38
Let
E = f (x)
_ : x 2 (0; 1)g :
Note that if (x) _ (x_ 0 ); then (x) _ 2 x_ 0 ; whence x0 (x) _ x0 x and
_ 2 x;
0
therefore x_ = x_ . Now let r1 ; r2 ; : : : be a one-one enumeration of Q \ ( 1; 1) ;
and for each n dene
En = E + rn = fy + rn : y 2 Eg :
We prove that
y + rm = x = z + rn :
6 The Lp Spaces
In this section we introduce certain innite-dimensional Banach spaces of sum-
mable functions that appear very frequently in many areas of pure and applied
mathematics. For convenience, we call real numbers p; q conjugate exponents
if p > 1; q > 1; and 1=p + 1=q = 1:
We begin our discussion with an elementary lemma.
39
Lemma 29 If x; y are positive numbers and 0 < < 1; then
x y1 6 x + (1 )y:
Proof. Taking u = x=y; consider
f (u) = u u 1+ :
0 1
We have f (u) = (u 1); which is positive if 0 < u < 1 and negative if
u > 1: Since f (1) = 0; it follows that f (u) 6 0 for all u > 0: This immediately
leads to the desired inequality.
holds.
R p p
Proof. We rst note that if jf j = 0; then jf j = 0 almost everywhere;
so
R f = 0; and therefore f g = 0, almost everywhere. Then f g is summable,
R q
f g = 0; and (1) holds trivially,
R as it does Ralso in the case where jgj = 0:
p q
Thus we may assume that jf j > 0 and jgj > 0: We then have, almost
everywhere,
p 1=p q 1=q
jf gj jf j jgj
R = R R q
p 1=p R q 1=q jf j
p
jgj
jf j jgj
p q
jf j jgj
6 R p + R q
p jf j q jgj
(where the last step uses Lemma 29), so
Z 1=p Z 1=q p q
jf j jgj
jf gj 6
p q
jf j jgj R p + R q : (2)
p jf j q jgj
Now, f g is measurable and the right-hand side of (2) is summable. Hence, by
Theorem 26, f g is summable and
Z Z Z 1=p Z 1=q
1 1
f g 6 jf gj 6
p q
jf j jgj + ;
p q
from which (1) follows.
Proposition 31 Let p > 1; and let f; g be measurable functions on R such that
p p p
jf j and jgj are summable. Then jf + gj is summable, and Minkowskis
inequality
Z 1=p Z 1=p Z 1=p
6
p p p
jf + gj jf j + jgj
holds.
40
p
Proof. Clearly, we may assume that p > 1: Now, jf + gj is measurable, by
Exercise (2.3.3: 5). Since
p
and the last function is summable, it follows from Theorem 26 that jf + gj
p 1
is summable. The functions jf j and jf + gj are measurable, by Exercise
(5.2, 4), and
q
p 1 p
jf + gj = jf + gj 2 L1 (R):
p 1
Thus, by Lemma 30, jf + gj jf j is summable and
Z Z 1 p 1 Z 1=p
jf j 6
p 1 p p
jf + gj jf + gj jf j :
p 1
Similarly, jf + gj jgj is summable and
Z Z 1 p 1 Z 1=p
jgj 6
p 1 p p
jf + gj jf + gj jgj :
It follows that
Z Z
p p 1
jf + gj = jf + gj jf + gj
Z Z
6 jf + gj
p 1 p 1
jf j + jf + gj jgj
Z 1 Z Z !
1 p 1=p 1=p
6
p p p
jf + gj jf j + jgj ;
Exercises (6.1)
N N
!1=p N
!1=q
X X X
xn yn 6
p q
jxn j jyn j
n=1 n=1 n=1
N
!1=p N
!1=p N
!1=p
X X X
6
p p p
jxn + yn j jxn j + jyn j
n=1 n=1 n=1
41
.2 A sequence (xn ) of real numbers is called p-power summable if the series
P
1
p
jxn j converges. Prove that if (xn ) is p-power summable and (yn ) is
n=1
q-power summable, where p; q are conjugate exponents, then
P
1
(i) xn yn is absolutely convergent, and
n=1
(ii) Hlders inequality holds in the form
1 1
!1=p 1
!1=q
X X X
xn yn 6
p q
jxn j jyn j :
n=1 n=1 n=1
Prove also that if (xn ) and (yn ) are both p-power summable, then so is
(xn + yn ) ; and Minkowskis inequality
1
!1=p 1
!1=p 1
!1=p
X X X
6
p p p
jxn + yn j jxn j + jyn j
n=1 n=1 n=1
holds.
.3 Let p > 1; and let lp denote the set of all p-power summable sequences,
taken with termwise addition and multiplication-by-scalars. Prove that
1
!1=p
X p
(xn )n>1 = jxn j
p
n=1
denes a norm on lp : (We dene the normed space lp (C) of p-power sum-
mable sequences of complex numbers in the obvious analogous way.)
is a norm, called the Lp -norm, on Lp (X) : the only property of a norm that
requires nontrivial verication is the triangle inequality, which in this case is
Minkowskis inequality, disposed of in Proposition 31.
When X = [a; b] is a compact interval, we write Lp [a; b] rather than Lp ([a; b]) :
Exercises (6.2)
In these exercises, X is a measurable subset of R:
42
.1 Let X be summable and 1 6 r < s: Prove the following.
r
(i) Ls (X) Lr (X). (Note that if f 2 Ls (X); then jf j 2 Ls=r (X):)
(ii) The linear mapping f f of Ls (X) into Lr (X) is bounded and has
1
s 1
norm 6 (X)r :
.2 Let 1 6 r 6 t 6 s < 1, r 6= s;
1 1 1 1
t s r t
= 1 1
; = 1 1
:
r s r s
and f 2 Lr (X) \ Ls (X): Prove that f 2 Lt (X) and
kf kt 6 kf kr kf ks :
t t
(Consider jf j jf j :)
.3 Prove that the step functions that vanish outside X form a dense subspace
of Lp (X) for p > 1: (First consider the case where X is a compact interval.)
.4 Clarksons inequalities: Let p; q be conjugate exponents, and let f; g 2
Lp (X): Prove that if 1 < p < 2; then
q 1
> kf + gkp + kf
p p q q
2 kf kp + kgkp gkp
and
p 1
gkp > 2 kf kp + kgkp
p p q q
kf + gkp + kf ;
Proof. We illustrate the proof with the case X = R and p > 1: Given a
Cauchy sequence (fn ) in Lp (R), choose a subsequence (fnk )k>1 such that
kfm fn kp 6 2 k
(m; n > nk ):
Then
fnk+1 fnk p
62 k
:
43
Writing q = p=(p 1); we see from Lemma 30 that for each positive integer N;
fnk+1 fnk is summable over [ N; N ]; and
Z Z 1=q
fnk+1 fnk [ N;N ] 6 fnk+1 fnk p [ N;N ]
62 k
(2N )1=q ;
so the series
1 Z
X
fnk+1 fnk [ N;N ]
k=1
converges. It follows from Lebesgues series theorem that there exists a null set
EN such that the series
1
X
fnk+1 (x) fnk (x) [ N;N ] (x)
k=1
P
1
converges for all x 2 RnEN ; and the function fnk+1 fnk [ N;N ] is sum-
k=1
mable. Then
1
[
E= EN
N =1
p
for all k; we see from Fatous lemma (Exercise (2.6, 5)) that jf j is summable
and hence that f 2 Lp (R). Moreover, if n > ni ; then by applying Fatous lemma
1
to the sequence (jfnk fn j)k=i we see that
kf fn kp = lim kfnk fn kp 6 2 i :
k!1
Hence limn!1 kf fn kp = 0:
Exercises (6.3)
44
7 Double Integrals
We next illustrate multiple Lebesgue integration by considering double integrals
By an interval in R2 we mean the Cartesian product I J of two intervals
in R: We say that f : R2 ! R is astep function if there exists a nite set of
non-overlapping bounded intervals Ik Jk (k = 1; ; n) such that
S
n
B f (x) = 0 for all x in the complement of Ik Jk ; and
k=1
where, we recall, jIk j denote the length of the interval Ik : Many fact, such
as the following, about the double integral of a step function are established
analogously to the corresponding facts for integrals of step functions on R; in
such cases we state the facts without proof.
2
Proposition 33 Let f and g be two stepR functions
R R Ron R such that f (x) = g(x)
for all but nitely many points x: Then f= g:
f + = max ff; 0g ;
and
f = max f f; 0g
are step functions. R R if f (x) 6 g(x) for all but nitely many (possibly no)
R R Finally,
points x, then f6 g:
45
RR
Then limn!1 fn = 0:
2
As in the one-variable case, this is a good
R Rdenition; L R is a linear 2space un-
der pointwise algebraic operations, and is a linear mapping of L R into R;
if f; g are summable functions dened almost R everywhere
R on R2 , then max ff; gg,
min ff; gg ; and jf j are summable, and f 6 jf j ; and the Lebesgue dou-
ble integral is translation invariant, in the sense that if f 2 L R2 and
yR R2 R2 ; then
R R the function fy dened by fy (x) = f (x + y) is summable, and
fy = f:
We say that f 2 L R2 is summable over the subset A of R2 if the
function f A belongs to L R2 ; we then write
Z Z Z Z
f= f A:
A
46
series theorem, Beppo Levis Theorem, Fatous lemma, Lebesgues dominated
convergence theorem, ... We can also dene measurability of functions dened
almost everywhere on R2 and of subsets of R2 ; and then prove analogues of
Propositions 2528.
Our main aim in this section is to deal with the relation between double
integration and repeated integration, familiar from second-year calculus courses.
First, given a two-variable function f; we dene a one-variable function fx :
R!R by fx (y) = f (x; y); likewise, if (fn ) is a sequence of two-variable functions,
we dene the corresponding one-variable functions fn;x by fn;x (y) = fn (x; y):
Proof. The general case easily follows from that in which f is a constant c
on a product I J of open intervals in R; and R is 0 elsewhere. Taking that case,
for each x 2
= I we have fx = 0 and therefore fx = 0; while for each x 2 I;
8
< c if y 2 J
fx (y) =
:
0 if y 2=J
R
and so, by denition of the integral on R; fx = c jJj : Thus (i) holds, g is well
dened, and 8
< c jJj if x 2 I
g(x) =
:
0 if x 2=I
Clearly, g is a step function on R; and
Z Z Z
g = c jJj jIj = f;
A(x) = fy 2 R : (x; y) 2 Ag
is a null set in R:
47
B f1 6 f2 6 ;
RR
B fn 6 M for each n; and
B fn (x; y) ! 1 for all (x; y) 2 A:
It follows from the rst and third of these properties that for each x 2 R the
sequence (fn;x )n>1 is increasing, and fn;x (y) ! 1 for all y 2 A(x): On the
other hand, by Lemma 37, fn;x is a step function on R; the function
Z
gn : x fn;x
R RR
is a step function on R; and gn = fn 6 M: Thus (gn )n>1 is a Riesz
sequence of step functions on R: Hence there exists a null subset B of R such
that (gn (x))n>1 converges for all x 2 RnB: In particular, for each such x there
R
exists Mx > 0 such that fn;x 6 Mx for all n: Thus (fn;x )n>1 is a Riesz
sequence of step functions on R and therefore converges almost everywhere in
R: Since this sequence diverges on A(x); we conclude that A(x) is a null set.
48
This proves (i). Moreover, we have
Z Z Z Z Z Z
fx = g = lim gn = f;
n!1
Proof. Exercise.
and dene
fn = min ff; ng :
49
Now, fn (x; y) ! f (x; y) for all (x; y) in the domain of f; the complement of
a set of measure zero: So, by Beppo Levis Theorem, f is summable. Finally, by
Fubinis theorem, there exist a summable function h on R and R a null set C R
such
R that
RR for each x 2 RnC; fx is summable and h(x) = fx ; and such that
h= f ; whence h(x) = g(x) for all x 2 Rn (S [ C) and
Z Z Z Z
g= h= f:
Exercises (7.1)
.1 Prove Corollary 40.
.2 Let f be a measurable function on R2 for which there exist a summable
functionR g on R and a null set A R such that jfx j is summable and
g(x) = jfx j for each x 2 RnA: Prove that f is summable.
.3 For each 2 R the function f : R2 ! R is dened by
8
< x (1 xy) if 0 6 x < 1; 0 6 y < 1
f (x; y) =
:
0 otherwise.
R
Find the values of for which f is summable, and evaluate f for those
values.
50
Exercises (8.1)
P
1
.1 Show that for each A R; (A) is the inmum of jIn j taken over all
n=1
covers of A by sequences (In )n>1 of bounded, but not necessarily open,
intervals.
.2 Prove that if a subset A of R has nite outer measure, then for each " > 0
there exists a sequence (In ) of disjoint bounded open intervals such that
[ P
1
A In and jIn j < (A) + ": (Recall that a nonempty subset of
n>1 n=1
.10 Prove that a subset A of R has nite outer measure if and only if l =
limn!1 (A \ [ n; n]) exists, in which case (A) = l:
.11 Prove that is translation invariant that is, (A + t) = (A) for each
A R and each t 2 R, where A + t = fx + t : x 2 Ag:
51
Proposition 42 The outer measure of any interval in R equals the length of
the interval.
Proof. Consider, to begin with, a bounded closed interval [a; b]: For each
" > 0 we have [a; b] (a "; b + ") and therefore
([a; b]) 6 j(a "; b + ")j = b a + 2":
Since " > 0 is arbitrary, we conclude that ([a; b]) 6 b a: To prove the reverse
inequality, let (In ) be any sequence of bounded open intervals that covers [a; b]:
Applying the HeineBorelLebesgue Theorem (1.4.6), and re-indexing the terms
In (which we can do without loss of generality), we may assume that for some
N;
[a; b] I1 [ I2 [ [ IN :
There exists an interval Ik1 , where 1 6 k1 6 N; that contains a; let this interval
be (a1 ; b1 ): Either b < b1 ; in which case we stop the procedure, or else b1 6 b:
In the latter case, b1 2 [a; b]n(a1 ; b1 ); so there exists an interval Ik2 ; where 1 6
k2 6 N and k2 6= k1 ; that contains b1 ; call this interval (a2 ; b2 ): Repeating this
argument, we obtain intervals (a1 ; b1 ); (a2 ; b2 ); : : : in the collection fI1 ; : : : ; IN g
such that for each i; ai < bi 1 < bi : This procedure must terminate with the
construction of (aj ; bj ) for some j 6 N: Then b 2 (aj ; bj ); so
N
X j
X
jIn j > (bi ai )
n=1 i=1
= bj (aj bj 1 ) (aj 1 bj 2)
(a2 b1 ) a1
> bj a1 :
P
1
It follows that jIn j > b a and therefore, since (In ) was any sequence of
n=1
bounded open intervals covering [a; b]; that ([a; b]) > b a: Coupled with the
reverse inequality already established, this proves that ([a; b]) = b a: The
proof for other types of interval is left as the next exercise.
Exercises (8.2)
.1 Complete the proof of Proposition 42 in the remaining cases.
.2 Let fI1 ; : : : ; IN g be a nite set of bounded open intervals covering Q\[0; 1]:
PN
Prove that jIn j > 1: (Given " > 0; extend each In , if necessary, to en-
n=1
sure that it has rational endpoints and that the total length of the intervals
is increased by at most ": Then argue as in the proof of Proposition 42.)
.3 Let X be a subset of R with nite outer measure. Prove that for each
" > 0 there exists an open set A X with nite outer measure, such
that (A) < (X) + ": (Use Exercise (8.1, 1).) Show that if X is also
bounded, then we can choose A to be bounded.
52
.4 Let A; B be subsets of R with nite outer measure. Prove that (A [ B) 6
(A) + (B):
We postpone the proof of this very useful theorem until we have dealt with
some auxiliary exercises.
Exercises (8.3)
We are now able to give the proof of the Vitali covering theorem.
53
(i) If n = 0; then
( n[1
)
1
jIn j > sup jIj : I 2 V and I \ Ik = ? ;
2
k=1
n[1
(ii) If n = 1; then In = In 1 and X Ik :
k=1
For each n > N let xn be the midpoint of In ; and let Jn be the closed interval
with midpoint xn and length 5 jIn j : It su ces to prove that
N
[ 1
[
Xn In Jn : (3)
n=1 n=N +1
For then
N
! 1 1
[ X X
Xn In 6 jJn j = 5 jIn j < ":
n=1 n=N +1 n=N +1
54
S
N
To prove (3), consider any x 2 Xn In : By Exercise (8.3,2), there exists
n=1
S
N
J 2 V such that x 2 J and J \ In = ?: We claim that J \ Im is nonempty
n=1
for some m > N: If this were not the case, then for each m we would have
S
m
J\ In = ? and therefore
n=1
( m
)
[
jJj 6 sup jIj : I 2 V; I \ Ik = ?
k=1
< 2 jIm+1 j ! 0 as m ! 1:
P
1
(For the last step, recall that jIm j is convergent.) It would then follow that
m=1
jJj = 0; which is absurd as V contains only non-degenerate intervals. Thus
= minfm > N : J \ Im 6= ?g
S1
is well dened, J \ In = ?; and therefore
n=1
( )
[1
jJj 6 sup jIj : I 2 V; I \ Ik = ? < 2 jI j :
k=1
jx x j 6 jJj + 1
2 jI j < 2 jI j + 1
2 jI j = 5
2 jI j :
In the remainder of this section we apply the Vitali covering theorem in the
proofs of some fundamental results in the theory of dierentiation and integra-
tion.
Let I be an interval in R. We say that a mapping f : I ! R is absolutely
n
continuous if for each " > 0 there exists > 0 such that if ([ak ; bk ])k=1 is a nite
Pn
family of non-overlapping compact subintervals of I such that (bk ak ) < ;
k=1
P
n
then jf (bk ) f (ak )j < ": On the other hand, if a; b 2 I and a < b; then we
k=1
dene the variation of f on the interval [a; b] to be
(n 1 )
X
Tf (a; b) = sup jf (xi+1 ) f (xi )j : a = x0 6 x1 6 6 xn = b ;
i=0
if this quantity exists as a real number; in that case we say that f has bounded
variation on [a; b] :
55
Exercises (8.4)
Prove that f is dierentiable at each point of [0; 1] but does not have
bounded variation on that interval.
.6 Prove that f : [a; b] ! R has bounded variation if and only if it can be
expressed as the dierence of two increasing functions. (For only ifnote
the last part of Exercise (8.4, 4).)
.7 Let f be absolutely continuous on a compact interval I = [a; b]: Prove
that f has bounded variation in I; that the variation function Tf (a; ) is
absolutely continuous on I; and that f is the dierence of two absolutely
continuous, increasing functions on I:
A simple corollary of the mean value theorem, one that su ces for many
applications, states that if f is continuous on [a; b] and jf 0 (x)j 6 M for all
x 2 (a; b); then jf (b) f (a)j 6 M (b a). Our rst application of the Vitali
covering theorem generalises this corollary, and can be regarded an extension of
the mean value theorem itself.
56
Proof. Let E I be a set of measure zero such that jf 0 (x)j 6 F 0 (x) for
each x 2 X = InE: We may assume without loss of generality that a; b 2 E:
Given " > 0; choose > 0 as in the denition of absolute continuity. For each
x 2 X there exist arbitrarily small r > 0 such that [x; x + r] (a; b),
and therefore
The sets of the form [x; x + r] ; for such r > 0; form a Vitali covering of X:
By the Vitali covering theorem, there exists a nite, pairwise disjoint collection
N
([xk ; xk + rk ])k=1 of sets of this type such that
N
!
[
Xn [xk ; xk + rk ] < :
k=1
and therefore
N
X1
jf (x1 ) f (a)j + jf (xk+1 ) f (xk + rk )j + jf (b) f (xN + rN )j < ":
k=1
It follows that
N
X1
jf (b) f (a)j 6 jf (x1 ) f (a)j + jf (xk+1 ) f (xk + rk )j
k=1
N
X
+ jf (b) f (xN + rN )j + jf (xk + rk ) f (xk )j
k=1
57
N
X
<"+ (F (xk + rk ) F (xk ) + 2"rk )
k=1
N
X1
6 " + F (x1 ) F (a) + (F (xk+1 ) F (xk + rk ))
k=1
N
X
+ (F (xk + rk ) F (xk ) + 2"rk )
k=1
+ (F (b) F (xN + rN ))
N
X
= " + F (b) F (a) + 2" rk
k=1
< F (b) F (a) + "(1 + 2b 2a):
Exercises (8.5)
for all 2 [a; x); and let s = sup X: Assume that s < b; and derive a
contradiction.)
.4 Let f be continuous on I = [a; b] ; and suppose there exists a countable
subset D of I such that f 0 (x) = 0 for all x 2 InD: Prove that f is constant
on I:
.5 Let C be the Cantor set. Show that [0; 1]nC is a countable union of non-
overlapping open intervals (Jn )n>1 whose lengths sum to 1; and that C
has measure zero.
P1 P1
For each x = an 3 n 2 C dene F (x)= an 2 n 1 : Show that
n=1 n=1
58
(i) if x has two ternary expansions, then they produce the same value
for F (x), so that F is a function on C;
(ii) F is a strictly increasing, continuous mapping of C onto [0; 1] ;
(iii) C is uncountable; and
(iv) F extends to an increasing continuous mapping that is constant
on each Jn ; equals 0 throughout ( 1; 0]; and equals 1 throughout
[1; 1):
Prove that for each > 0 there exist nitely many points
N
X
(F (bn ) F (an )) = 1;
n=1
P
N
and (bn an ) < : (Thus F is increasing and continuous, but not
n=1
absolutely continuous, on [ 1; 2] :)
Finally, show that F 0 (x) = 0 for all x 2 [0; 1]nC; but F (1) > F (0):
59
or else (f (x + h) f (x)) /h remains unbounded as h ! 0+ :
Exercises (8.6)
.1 Prove that D+ f (x) > D+ f (x) and D f (x) > D f (x) whenever the
quantities concerned make sense.
.2 Prove that f is dierentiable on the right (respectively, left) at x if and
only if D+ f (x) = D+ f (x) (respectively, D f (x) = D f (x)):
and
m
X
(f (xi ) f (xi hi )) < s ( (E) + ") :
i=1
Again applying the Vitali covering theorem, prove that there exist nitely
S
m
many points y1 ; : : : ; yn of E \ Ji ; and nitely many positive numbers
i=1
h01 ; : : : ; h0n ; such that
60
Our next theorem shows, in particular, that the dierentiability of the func-
tion F can be dropped from the hypotheses of Proposition 44.
Proof. To prove that F is dierentiable a.e., it su ces to show that the sets
S = x 2 R : D+ F (x) is undened ;
T = x 2 R : D+ F (x) > D F (x)
have measure zero. For, applying this and Exercise (8.6, 3) to the increasing
function x F ( x); we then see that D F (x) 6 D+ F (x) almost everywhere;
whence, by Exercise (8.6, 1)
where a < b and r; s are rational numbers with r > s, it is enough to prove that
such a set E has measure zero. Accordingly, x " > 0 and use Exercise (8.6, 5)
to obtain
(i) nitely many points x1 ; : : : ; xm of (a; b) ; and nitely many positive num-
bers h1 ; : : : ; hm ; such that the intervals Ji = (xi hi ; xi ) (1 6 i 6 m)
form a pairwise disjoint collection,
m
!
[
Ji > (E) ";
i=1
and
m
X
(F (xi ) F (xi hi )) < s ( (E) + ") ;
i=1
61
S
m
(ii) nitely many points y1 ; : : : ; yn of E \ Ji ; and nitely many positive
i=1
numbers h01 ; : : : ; h0n ; such that
Si = fk : (yk ; yk + h0k ) Ji g :
so that
s( (E) + ") > r ( (E) 2") :
Since " > 0 is arbitrary, it follows that s (E) > r (E): But r > s; so we must
have (E) = 0:
For each n write
1
Fn (x) = F x+ (x 2 R)
n
and
fn = n (Fn F):
Note that, being an increasing function, F is Riemann integrable, and therefore
summable, over any proper compact interval [a; b] in R; whence F; Fn and fn
are summable over [a; b] : Let a 6 < b; and choose a positive integer N such
that + N1 < b: For all n > N we have
Z Z Z
fn = n Fn n F
a a a
Z 1
+n Z
=n F n F
1
a+ n a
Z 1
+n Z 1
a+ n
=n F n F;
a
62
where we have used Exercise (2.3, 4) to produce the second-last line. Now, F is
increasing and + n1 < b; so
Z 1
+n Z 1
+n
n F 6n F (b) = F (b):
It follows that Z
fn 6 F (b) F (a) (n > N ) :
a
Next we note that since fn (x) ! F 0 (x) almost everywhere, F 0 is measurable.
Also, fn > 0; so Fatous lemma (Exercise (2.6, 5)) can be applied, to show that
F 0 is summable over [a; ] and
Z Z
F 6 lim inf
0
fn 6 F (b) F (a):
a a
Since
F 0 [a;b 1
] (x) ! F
0
[a;b] (x) as k ! 1
k
for all x 2 [a; b]; and, as is easily conrmed, F 0 [a;b 1 ] is a Riesz sequence
k k>1
of summable functions, we now see from Beppo Levis Theorem that
Z b Z Z b 1
n
F 0 = lim F 0 [a;b = lim F 0 6 F (b) F (a);
n]
1
a n!1 n!1 a
as we wanted.
Proof. Fix real numbers a; b with a < b: It su ces to prove that F 0 (x) =
P
1
Fn0 (x) almost everywhere on I = [a; b]: for then we can apply the result to
n=1
the intervals [ n; n] as n increases through N+ . If necessary replacing Fn by
Fn Fn (a); we may assume that Fn (a) = 0: Write
63
P1
and note that F sn = k=n+1 Fk is increasing and nonnegative on I. By
Theorem 45, sn is dierentiable on InAn for some set An of measure zero;
likewise, F (which is clearly increasing) is dierentiable on InA0 for some set
A0 of measure zero. Then
1
[
A= An
n=0
has measure zero. Since both F sn+1 and sn+1 sn are increasing functions,
for each x 2 InA we have
P
1
It follows from the monotone sequence principle that Fn0 (x) converges to a
n=1
sum 6 F 0 (x):
Now choose an increasing sequence (nk )k>1 of positive integers such that for
each k;
0 6 F (b) snk (b) 6 2 k :
Since F snk is an increasing function, for each x 2 I we obtain the inequalities
almost everywhere on I:
Exercises (8.7)
.1 Let f be an increasing function on [a; b], and for each positive integer n
dene
Sn = x 2 (a; b) : D+ f (x) > n :
64
Prove that
1
(InSn ) < n (f (b) f (a))
and hence that the set of those x 2 (a; b) at which D+ f (x) is undened
has measure zero. (Use the Vitali covering theorem to show that there
exist nitely many points x1 ; x2 ; : : : ; xm of (a; b); and positive numbers
h1 ; h2 ; : : : ; hm ; such that xk + hk < xk+1 and f (xk + hk ) f (xk ) > nhk :)
.2 Let E be a bounded subset of R that has measure zero, and let a be a
lower bound for E: For each positive integer n choose a bounded open set
An E such that (An ) < 2 n ; and dene
8
< 0 if x < a
fn (x) =
:
(An \ [a; x]) if x > a:
Show that
P
1
(i) f = fn is an increasing continuous function on R;
n=1
(ii) D+ f (x) is undened for each x 2 E:
The sets [x; x + r] of this type form a Vitali cover of X: With the aid
of the Vitali covering theorem, construct a partition P of I such that
U (P; f ) L(P; f ) < ": This provides an alternative proof of one half of
Theorem 23.
65
In order to prove this theorem, we rst prove it in the case where f is
bounded.
1
Fn (x) = F x+
n
and Z 1
x+ n
fn (x) = n (Fn (x) F (x)) = n f:
x
Note that Z 1
x+ n
fn (x) 6 n c=c
x
for all x in the domain of f: Now, by Exercise (2.7, 5), F is continuous; so both
Fn and fn are continuous and therefore, by Exercise (5.1, 3), measurable. For
all a; b with a < b; since fn (x) ! F 0 (x) almost everywhere on [a; b], we see from
Lebesgues dominated convergence theorem that F 0 [a;b] is summable and that
for each x 2 [a; b] ;
Z x Z x
F 0 = lim fn
a n!1 a
Z x
1
= lim n t F t+ F (t)
n!1 a n
Z 1 Z !
x+ n x
= lim n F F
n!1 1 1
Z 1
x+ n
= lim n F
n!1 x
= F (x);
where the second last step uses translation invariance (Exercise (2.3, 4)) and the
nal line follows from Exercise (4.1, 3). Hence
Z x
(F 0 f ) = 0 (x 2 [a; b]) :
a
It follows from Exercise (2.7, 6) that F 0 (x) = f (x) almost everywhere on [a; b] :
A by-now-standard argument shows that F 0 (x) = f (x) almost everywhere, and
completes the proof.
66
We now give the proof of the fundamental theorem of calculus.
Proof. Under the hypotheses of Theorem 47, we may assume without loss
of generality that f > 0: For each positive integer n and each x 2 R dene
Z x
Fn (x) = (f min ff; ng) :
1
Hence
Z x
d d
F 0 (x) = Fn + min ff; ng > min ff (x); ng a:e:
dx dx 1
Thus Z
(F 0 f) [ k;k] = 0:
Proposition 49 Let f be a summable function, and let a < b: Then the indef-
inite integral Z x
F (x) = f
a
67
a nite collection of non-overlapping compact subintervals of [a; b] such that
[n
Pn
(bk ak ) < ; and let A = [ak ; bk ] : Then (A) < ; so
k=1 k=1
n
X n Z
X bk
jF (bk ) F (ak )j = f
k=1 k=1 ak
n
X bkZ Z
6 jf j = jf j < ":
k=1 ak A
Conversely, we have
Proposition 50 Let F be an absolutely continuous mapping on a compact in-
terval I = [a; b] : Then there exists a summable function f such that
Z x
F (x) F (a) = f (x 2 I) :
a
Let Z x
G(x) = F0 (a 6 x 6 b) :
a
Then G is absolutely continuous, by Proposition 49, as is the function F G: By
the fundamental theorem of calculus, F 0 (x) G0 (x) = 0 almost everywhere on
[a; b] : It follows from Exercise (8.5, 2) that F G is constant on [a; b] ; whence
Z x
F (x) F (a) = G(x) G(a) = F 0;
a
68
Appendix: The axiom of choice
In the early years of this century it was recognised that the following principle,
the axiom of choice, was necessary for the proofs of several important theorems
in mathematics.
In particular, Zermelo used this axiom explicitly in his proof that every set S
can be well ordered that is, there is a total partial order > on S with respect
to which every nonempty subset of X has a least element [13]. It was shown
by Gdel [5] in 1939 that the axiom of choice is consistent with the axioms of
ZermeloFraenkel set theory (ZF), in the sense that the axiom can be added
to ZF without leading to a contradiction, and by Cohen [4] in 1963 that the
negation of the axiom of choice is also consistent with ZF. Thus the axiom of
choice is independent of ZF: it can be neither proved nor disproved without
adding some extra principles to ZF.
The axiom of choice is commonly used in an equivalent form (the one we
used in the construction of a non-measurable set):
To prove the equivalence of these two forms of the axiom of choice, rst assume
that the original version AC of the axiom holds, and consider nonempty sets
A; B and a subset S of A B such that for each x 2 A there exists y 2 B with
(x; y) 2 S: For each x 2 A let
Fx = fxg fy 2 B : (x; y) 2 Sg :
Then F = (Fx )x2A is a nonempty family of pairwise disjoint sets, so, by AC,
there exists a set C that has exactly one element in common with each Fx : We
now dene the required choice function f : A ! B by setting
for each x 2 A:
Now assume that the alternative form AC0 of the axiom of choice holds, and
consider a nonempty family F of pairwise disjoint nonempty sets. Taking
A = F;
[
B= X;
X2F
S = f(X; x) : X 2 F; x 2 Xg
69
in AC0 , we obtain a function
[
f :F ! X
X2F
such that f (X) 2 X for each X 2 F. The range of f is then a set that has
exactly one element in common with each member of F:
There are two other choice principles that are widely used in analysis. The
rst of these, the principle of countable choice, is the case A = N of AC0 .
The second is the principle of dependent choice:
It is a good exercise to show that the axiom of choice entails the principle of
dependent choice, and that the principle of dependent choice entails the principle
of countable choice. Since the last two principles can be derived as consequences
of the axioms of ZF, they are denitely weaker than the axiom of choice.
For a fuller discussion of axioms of choice and related matters, see the article
by Jech on pages 345370 of [2].
70
References
[1] E. Asplund and L. Bungart, A First Course in Integration, Holt, Rinehart
& Winston, New York, 1966.
[4] P.J. Cohen, Set Theory and the Continuum Hypothesis, W.A. Benjamin,
Inc., New York, 1966.
[5] K. Gdel, The Consistency of the axiom of choice and the Generalized Con-
tinuum Hypothesis with the Axioms of Set Theory, Annals of Mathematics
Studies 3, Princeton University Press, Princeton, NJ, 1940.
[6] P.R. Halmos, Measure Theory, Graduate Texts in Mathematics 18,
Springer-Verlag, Heidelberg, 1974.
[7] H. Lebesgue, Intgrale, longueur, aire, Annali di Mat. (3) 7, 231259,
1902.
[8] H. Lebesgue, Leons sur lintgration et la recherche des fonctions primi-
tives, Gauthier-Villars, Paris, 1904.
[9] F. Riesz, Sur lintgrale de Lebesgue comme lopration inverse de la dri-
vation, Ann. Scuola Norm. Sup. Pisa (2)5, 191212, 1936.
[10] F. Riesz and B. Sz.-Nag, Functional Analysis, Frederick Ungar, New York,
1955; reprinted in paperback by Dover Publications Inc., New York, 1990.
[11] H.L. Royden, Real Analysis (3rd edn), Macmillan Publ. Co., New York,
1988.
[12] Walter Rudin, Real and Complex Analysis, McGraw-Hill, New York, 1970.
[13] E. Zermelo, Beweis, dass jede Menge wohlgeordnet werden kann, Math.
Annalen 59, 514516, 1904.
71